Checking the features:

MSE: 3.0550680492414715e+26

The matches were doing by categorizing the states based on the $\bar{x}\pm 2 \cdot sd(x)$ for each feature

The ideal accuracy is by setting thresholds with higher $sd$ and allowing it to meet higher criteria (ideally all)

The time stamps were categorised and the the forecasted values were compared against it [best so far]

Total number of states: 15805
Matched states : 11059; 	    Accuracy: 69.97%
Matched within 1 state: 11675; 	Accuracy within 1 state: 73.87%
unassigned : 3033,              Percentage : 19.19% 

The graph of the different features:

statsexchange_link